Some bugs found during 16hour debug session
 

After spamming server with ~1000connections/sec (where ~150 were accepted/sec), I found several bugs:

#1. Work thread count was not dynamic, as supposed to be
#2. Two pointers were uninitialized (work thread queue.. they seemed always to be NULL on program initialization.. so no harm done)
#3. Unregistered version was using Heap that was meant for registered version, which caused oddities under heavy load.
#4. Something is leaking handles.. :? Might be winsock issue, under heavy load (need to investigate it further).